VoxPoser: Composable 3D Value Maps for Robotic Manipulation with Language Models

by Stanford University · free · Last verified 2026-03-17

VoxPoser uses LLMs and vision-language models to synthesize 3D voxel-based value and constraint maps that guide robot motion planners, enabling zero-shot generalization to novel language instructions and object configurations. The approach produces trajectories without any robot-specific training by composing affordance maps in 3D space.